Stock Investment Cross-sectional Income Forecast ——A View Based On The Combined Model

Building a forecasting model that can predict stock price trends can help investors make the right decisions,improve profitability and reduce potential losses.Since there are many reasons affecting stock market volatility and the interactions among the factors are intricate,many finance scholars are trying to find the factors that affect stock returns,including company fundamentals,market environment,and policies.More and more scholars have found that existing forecasting models are not well adapted to the changing characteristics of stock markets and have many limitations.These limitations are reflected in the uncertainty of stock market forecasting model parameters,forecast instability,and noise accumulation.It is these shortcomings that make it difficult to accurately analyze and predict the trend of stock market changes,and it is difficult to identify one or one aspect of the factors that have a greater impact on the stock market.In this paper,the analysis of the existing literature concludes that the existing stock return forecasting models have the following three main problems:(1)Insufficient selection of feature variables to identify all factors affecting stock returns.(2)The introduction of too many characteristic variables in the model leads to a decrease in the predictive validity of the model.(3)The high correlation between the feature variables leads to the problem of unstable prediction.To solve the above three problems,this paper builds a suitable model based on the financial and stock trading data of listed companies from January 2010 to December 2020.Firstly,the traditional variable selection method is used for modeling,and then the upward and downward trends of stocks are predicted,and the prediction accuracy obtained is not satisfactory.Then this paper proposes two methods to build a new model based on the idea of a combination model,a combination Elastic Net,and combination Adaptive-Lasso,and tests the prediction performance of the model by prediction slope,prediction accuracy,and asset portfolio allocation.According to the test results,it can be obtained that: 1.the prediction results of the two models proposed in this paper show a positive relationship with the real rise and fall of stocks,and the prediction accuracy is higher.2.both models pass the ADF test,indicating that the models have good prediction stability.3.The predictive performance of the model does not change with changes in financial markets,again indicating that the model is more stable and belongs to a market-neutral investment strategy,which can achieve better predictive results in most markets.4.Asset portfolio allocation based on the model's predictive results can yield higher returns than traditional logistic models.This paper combines variable selection methods and portfolio forecasting ideas to build two models,Portfolio Elastic Net-logistic and Portfolio Adaptive Lassologistic,which have been tested to show the superiority of the models.The empirical results show that these two models can effectively identify the complex relationship between characteristic variables and expected returns,and have higher forecasting accuracy compared with the traditional multivariate logistic models and variable selection methods,which can bring more benefits to the majority of investors.It is also found that the firm characteristic variables affecting the expected returns of stocks in China's stock market are constantly changing over time,and the significant dynamic changes of these characteristic variables reflect some extent the weak stability of Chinese stock market.